测试结果

[qianmeng19]

我的测试结果如图，这种结果是不是无效的？

[iamstarlee]

是测试集的结果吗？测试集没有ground truth，只能用验证集验证测试两用

[qianmeng19]

是的是测试集结果，我使用Train13和Train13作为训练集和测试集，使用Xie2019模型，结果却如图，不应该是为0或者1才是对的嘛

[iamstarlee]

模型的输出结果会经过sigmoid激活到0～1之间，0或者1是真值，模型好的预测只会很靠近0或者1,几乎不会为0或者1

[qianmeng19]

那怎样的结果才是模型好的预测呢

[qianmeng19]

可以看看博主的测试结果吗

[iamstarlee]

Multi-label-Sewer-Classification/metrics.py

Lines 124 to 126 in cd4bd6b

	Ng[k] = np.sum(tmp_targets == 1)
	Np[k] = np.sum(tmp_scores >= threshold) # when >= 0 for the raw input, the sigmoid value will be >= 0.5
	Nc[k] = np.sum(tmp_targets * (tmp_scores >= threshold))

需要先用inference.py进行推理，然后用calculate_results.py计算推理结果，从而判断模型的好坏，metrics.py里计算指标的时候会把高于阈值的算为正样本，低于的算为负样本

[iamstarlee]

我在计算指标之前就用阈值将结果置为0或者1了

[qianmeng19]

感谢博主分享，请问怎样在计算指标之前就用阈值将结果置为0或者1，还有就是阈值一般设置为多少

[iamstarlee]

Multi-label-Sewer-Classification/metrics.py

Lines 99 to 129 in cd4bd6b

	def evaluation(scores, targets, weights, threshold = 0.5):
	assert scores.shape == targets.shape, "The input and targets do not have the same size: Input: {} - Targets: {}".format(scores.shape, targets.shape)
	_, n_class = scores.shape
	# Arrays to hold binary classification information, size n_class +1 to also hold the implicit normal class
	Nc = np.zeros(n_class+1) # Nc = Number of Correct Predictions - True positives
	Np = np.zeros(n_class+1) # Np = Total number of Predictions - True positives + False Positives
	Ng = np.zeros(n_class+1) # Ng = Total number of Ground Truth occurences
	# False Positives = Np - Nc
	# False Negatives = Ng - Nc
	# True Positives = Nc
	# True Negatives = n_examples - Np + (Ng - Nc)
	# Array to hold the average precision metric. only size n_class, since it is not possible to calculate for the implicit normal class
	ap = np.zeros(n_class)
	for k in range(n_class):
	tmp_scores = scores[:, k]
	tmp_targets = targets[:, k]
	tmp_targets[tmp_targets == -1] = 0 # Necessary if using MultiLabelSoftMarginLoss, instead of BCEWithLogitsLoss
	Ng[k] = np.sum(tmp_targets == 1)
	Np[k] = np.sum(tmp_scores >= threshold) # when >= 0 for the raw input, the sigmoid value will be >= 0.5
	Nc[k] = np.sum(tmp_targets * (tmp_scores >= threshold))
	ap[k] = average_precision(tmp_scores, tmp_targets)
	#print("the Ng is {}".format(Ng))

第125行，Np[k]是第k类的Total number of Predictions，这里就是在取预测值大于阈值（一般为0.5）的结果

[qianmeng19]

感谢博主的回答，但我想像博主一样将[metrics.py]中的评估函数运用到inference.py中进行推理，需要怎样修改代码呢

[iamstarlee]

运行inference.py就会得到推理结果，再运行calculate_results.py就能得到推理结果的指标，里面会调用metrics.py

[qianmeng19]

会不会出现inference.py就会得到推理结果很差，但是运行calculate_results.py得到的结果也还行。我想知道calculate_results.py的具体计算公式，论文里有吗

[qianmeng19]

请问博主是使用了所有数据集进行训练吗

[iamstarlee]

calculate_results.py是基于inference.py的结果计算的，calculate_results.py的计算公式都在metrics.py里，论文的附录里有

[iamstarlee]

我是用的整个数据集训练的

[qianmeng19]

请问博主试过只使用一个文件train13训练吗，我使用一个文件训练的结果好像不行

[qianmeng19]

请问博主执行calculate_results.py时遇到这个报错是结果有问题吗E:\Multi-label-Sewer-Classification-main\metrics.py:173: RuntimeWarning: invalid value encountered in scalar divide
F2_normal = (5 * precision_k[-1] * recall_k[-1])/(4*precision_k[-1] + recall_k[-1])

[iamstarlee]

这种看上去像除了0，可以试着print分母的precision和recall

[iamstarlee]

train13是我从原始数据集中摘出来的子集，可能因为数量少导致效果不好